KR20080024126A - Constrained exploration for search algorithms - Google Patents

Abstract

Partition criteria is used to direct or constrain a search process that searches a search space for solution to a problem. The identification of states for further exploration of the search space is directed by the use of reference state and distance constraint information that is part of the partition criteria. Multiple processes use the same or different partition criteria to together search relevant areas of a search space. The operation of multiple processes, including the choice of partition criteria, is coordinated by a coordination or control process. Different search topologies are implemented by selecting various partition criteria. ® KIPO & WIPO 2008

Description

Constrained investigation for search algorithms {CONSTRAINED EXPLORATION FOR SEARCH ALGORITHMS}

Many problems, including problems in planning, scheduling, and configuration, can be solved by a search algorithm, which is an algorithm that defines how to search a problem space or "search space" for a solution to a problem. For example, the problem of finding a suitable network topology for multiple computer systems with different attributes, different connections, etc. can be solved by a search algorithm. Determining a schedule that is suitable for a set of resources that can meet different needs is another problem and the use of a search algorithm can be useful for this.

In general, methods for solving these problems can be classified as deterministic and nondeterministic. The deterministic method includes a clear and ordered set of steps that can perform an efficient search of the search space to find a solution. On the other hand, a nondeterministic method examines the search space in a way that can depend on more than just input and current state, and the search path of this nondeterministic method cannot be absolutely predicted. For many problems, there is no efficient deterministic method, and the most likely way to solve a particular problem is to use a nondeterministic method.

For many of these problems, investigating the search space requires a large amount of resources to compute regardless of the nature of the search algorithm used. A common way of accelerating search is to apply a number of individual computing resources, including computer processors with multiple cores, computers with multiple processors, and computers configured in many different ways. Some deterministic algorithms are suitable for the use of multiple computing resources, and the search space can be divided into multiple independent subspaces, each of which is then examined by a separate computing resource. On the other hand, in many nondeterministic algorithms, it is difficult to efficiently utilize multiple computing resources in the same way. For example, separating search space before starting a non-deterministic search cannot guarantee investigative diversity, because predicting the final search path of many non-deterministic algorithms, i.e. given a starting search state This is because it may be difficult or impossible to determine which part of the search space to investigate.

The following description provides a brief overview of the specification for a basic understanding of those skilled in the art. This Summary is not an extensive overview of the specification and does not identify major / critical elements of the invention or limit the scope of the invention. Its sole purpose is to present some concepts described herein in a simplified form as a prelude to the more detailed description that is discussed later.

In this specification, various techniques and techniques for methods and systems for constraining and directing the investigation of search algorithms are described. Applying these methods and systems to search investigations results in more efficient use of multiple individual computing resources.

Exemplary embodiments described herein use multiple processes to search the space in which results are identified. In general, a search space is partitioned into regions using one or more partition criteria, which can be represented as a measure of distance around a reference search state and generally a reference search state to be searched by one or more search processes. Depending on how the partitioning criteria are selected and how they are assigned to the various search processes, the search topology and strategy may utilize various features and have overlapping search spaces, non-overlapping search spaces, or a combination of both. Can be. In dividing the search space, the processes may be coordinated by a coordination or control process, which may be a search process of any one of the search processes or may be a separate control process. Alternatively, retrieval processes can exchange information for coordination between themselves partitioning the search space. Exemplary embodiments described herein may utilize any type of numerical model to evaluate candidate search status, and to search status within a region until a suitable condition is found and / or until a predetermined interruption criterion is met. Any type of strategy can be used to select them.

1 illustrates a general operational flow, including data that may be used by a search algorithm that searches the search space and various operations that may be performed.

2 illustrates a general system using one or more instances of search algorithms to search a search space.

3 illustrates a general operational flow, including various operations that may be performed using search algorithms that search a search space without a centralized partitioning agent or logic.

4 illustrates a general representation of a system illustrating one of many possible search topologies by indicating splitting one search space into two search spaces.

5 illustrates a general representation of a system that illustrates how a search changes over time by using a new segmentation criterion.

6 illustrates a general representation of a system illustrating one of many possible cases of dividing one search space into three search spaces.

FIG. 7 illustrates a general representation of a system illustrating separation into two search spaces in which the reference states are not the same and the search spaces do not surround all of the entire search space.

8 illustrates a general representation of a system illustrating separation into two search spaces where the reference states are not the same and the final search spaces overlap.

9 illustrates a general representation of a system illustrating separation into three search spaces in which a third search space is defined, using partitioning criteria including reference conditions and distance constraints.

10 illustrates one possible basic implementation of a computing device.

Referring to FIG. 1, an exemplary and general operational flow 100 is shown that includes data that may be used by a search algorithm that searches a search space and various operations that may be performed. Given certain state data 110 as well as other data such as break criteria 122 and partition criteria 118, operation flow 100 examines the search space by evaluating the model and identifying new states. .

Although FIG. 1 may be described with reference to other figures, it is to be understood that the illustrative operational flow 100 is not intended to be limited to any particular figure or system of figures or other content. In addition, it is to be understood that although the example operational flow 100 indicates a particular order of action execution, in one or more other implementations the actions may have a different order. In addition, some of the data and steps illustrated in the example operational flow 100 may not be necessary and may be omitted in some implementations. Finally, while this flow of operations involves a number of individual steps, it should be appreciated that in some circumstances some of these operations may be combined and executed simultaneously.

In general, search algorithms try to find one or more solutions to a particular problem. Such algorithms generally operate on models. The model provides some form of representation that is used as part of problem solving. The model provides a result, given a predetermined state. This result can then be assessed if it is suitable for solving the problem, and some results will be better solutions to the problem than others. The search algorithm can then select new values for the state, using any number of other factors as well as the existing state, result, and suitability of the result. The search algorithm can then provide this new state to the model, get the results, evaluate the results, and so on. This process can continue until certain interruption criteria are met. For example, the process may continue until the completion of the inspection of the predetermined number of states, until the result meets a certain standard of conformity, or until some other criterion is met.

In some exemplary cases, the problem to be solved is in the physical domain, but is not limited to this example. In some of these cases, the model may be a representation of a given physical process or processes, and the state provided to this model may be the coding of physical parameters, and the result generated by this model may be the It may be the result of the manipulation. For example, suppose that the problem to be solved is related to configuring a computer server, but is not limited to this example. The state provided to the model in this example includes the number of servers available, the speed of the servers, the amount of memory used by the servers, the speed of the networks connecting the servers, and the complexity of the processing performed by the servers. It may include. The types of these inputs are represented by (A, B, C, D, E) and the specific set of state data is represented by (a, b, c, d, e), and each value is used to identify the solution. Assume that it corresponds to a variable in the model that is being modeled. In this example, the model can perform a variety of calculations, including an estimate of the average response time that each user experiences while using the computer servers, as well as an estimate of the concurrent users that the computer servers represented by a given state can support. It can provide results that include numbers. Assume that the model of the result is represented by (Y, Z) and the model provided with inputs such as (a, b, c, d, e) produces a result with outputs such as (y, z).

One of the factors that distinguishes different search algorithms is how the search algorithm identifies the next state or states to evaluate. Identification of the following states can be accomplished using many different methods, all or most of which may be useful in the context of the present invention. For example, some search algorithms may randomly identify the next state, while other search algorithms may include, for example, genetic algorithms, slopes of current and recent results, or some other method. By using this, it is possible to identify the next state from a family of states related to the current state in some way.

By using a "distance constraint" in determining the next state, one can constrain the search algorithm's investigation in a variety of beneficial ways. In general, a “distance constraint” can constrain the algorithm based on a predetermined measurement of distance or the difference between the two states. If the search algorithm can only identify subsequent states that meet the constraints specified by the distance constraint, then the search algorithm's investigation can be limited in various ways for various purposes.

In some example implementations, distance constraints or distance constraints may be used for multiple instances of one or more search algorithms, such that each instance of the search algorithm will examine a particular search space that is part of the overall search space. In such an example, depending on the nature or constraints of the states and distance constraints provided to the models, avoiding overlap between different instances of the search algorithm, the same portion of the search space is searched many times in the same manner. And certain areas of the search space may be examined in more detail than others.

As shown, one implementation of operational flow 100 utilizes state data 110. This state data 110 represents the current state as understood by the search algorithm. During the first iteration of the operational flow 100, the state data 110 may include an initial state provided to a search algorithm, or an initial state selected within a designated search space. As described in more detail below, during subsequent iterations, state data 110 may include a new state that is identified based on a previous iteration or execution of iterations.

In some example implementations, state data 110 may be represented by values of one or more variables, but is not limited to this example. As mentioned above, in this exemplary implementation, state data may be represented as (a, b, c, d, e) and may indicate things such as the number of available computer servers, the speed of these servers, and the like. . It will be appreciated that the number and composition of variables, as well as the data represented by the variables, as well as the use of all variables generally have a problem to solve and the model and functional steps used in the solution of that problem. Representation of states is not critical in the context of the present invention, and state data 110 may be represented in any manner.

In one implementation of operation 130, the search algorithm evaluates the state data 110 using a particular model related to the problem being investigated by the search algorithm. The model provides some form of representation that is used as part of problem solving. The model provides a result, given a predetermined state. The model can be implemented using any of a variety of methods and can be changed depending on the nature of the problem being solved.

The result of evaluating the model given the state data 110 is illustrated by the result data 112 in FIG. 1. In some example implementations, result data 112 is typically, but not always, represented as values for one or more variables that represent the parameters examined to determine suitability or adaptability of the state in producing a desired result. It may be, but is not limited to such. In the example described above, these variables may indicate such as the number of concurrent users that a set of computer servers can support and the average response time experienced by each user of the modeled system. As with the state data 110, the use of all variables, as well as the number and configuration of the variables, the data represented by the variables, are specific details of a particular implementation of the search algorithm investigation, and the resulting data 112 is arbitrary It will be appreciated that it may be expressed in a manner.

In one implementation of operation 150, operation flow 100 determines whether abort criteria 122 are met. If it is determined in operation 150 that the abort criteria are met (“YES” branch, operation 150), the flow of operations proceeds to operation 170 as described below. If it is determined that an abort criterion is not met (“No” branch, operation 150), the operation flow proceeds to operation 160 as described below.

The abort criteria 122 can be any criteria that specifies when the search should end. For example, the stop criteria 122 may include a temporal criterion, such as but not limited to, the time at which the search can be executed or the number of repetitions. Alternatively, other criteria may be used, such as a measure of suitability or adaptability. Suspension criteria may specify that by using the above example, the number of concurrent users that can be supported by a set of computer servers exceeds a certain amount, or that the response time is less than a certain specified time. Can be. Any other criteria as desired may be used, including criteria for which a suitable solution has been found by other currently running search processes. The possibilities for the break criteria 122 are substantially endless.

Once the abort criteria 122 have been met, the operational flow 100 proceeds to an operation 170 in which the operational flow 100 returns a predetermined search result in one implementation. The search results may include, but are not limited to, the most recently processed state data 110. This search result may or may not include the most recent result data 112. In some implementations, the search results can include more than one portion of the state data 110, the result data 112, or other data. In some implementations that return more data than a set of data, some or all of the data may be stored in a data store, such as data store 180, at or between longer periods of time. In addition, in some cases, operation flow 100 may not return any value. For example, if the operation flow is executed a certain number of times without finding result data 112 that meets certain specific criteria, operation 170 may not return any results, but is not limited to this example. Another example where operation 170 may not return any results is, in some implementations, where the operation flow is interrupted because another instance of the operation flow has found a solution. Alternatively, in other implementations, operation 170 may return data that most closely matches a predetermined criterion, or some other set of data that may be useful in some manner to a user of a search algorithm. .

If the abort criterion 122 is not met, the operational flow 100 proceeds to operation 160. In one implementation of operation 160, operation flow 100 identifies a new state to be used in its subsequent iteration. The operation flow proceeds back to operation 130 using the state identified in operation 160 after completing execution of operation 160.

The way of identifying a new state by operation 160 can vary widely. In some implementations, the operational flow can identify a new state through a process that it executes as part of operation 160. For example, operation 160 may identify a new state using an algorithm, but is not limited to this example, in situations where data such as an existing state and suitability of an existing state is given. In the same or other implementations, operation 160 may identify the new state by, for example, obtaining a new state from an input outside the scope of any other process or operation 160. For example, any other process, including another search process, may identify a new state to be used by operation 160, but is not limited to this example.

Depending on the particular search algorithm and other choices used, the details of how the next state is identified can be changed drastically. For example, in some implementations, the next state can be randomly identified. In other implementations, the next state can be identified from a family of states that are related in some way to the current state, for example by using genetic algorithms, slopes of current and recent results, or some other method. In the context of the present invention, any such algorithm or process may be used to identify a new state.

In one particular implementation, the next state may be constrained using any particular algorithm by itself or in addition to other constraints or criteria, and by partition criteria 118. Partition criteria 118 may include both “distance constraint” and “reference state”. Depending on the distance constraint and the choice of reference state, many different search schemes can be obtained. Some examples of search schemes that may arise from different selections for splitting reference parameters are described below with reference to FIGS. 4-9.

Depending on the exact embodiment and implementation, the partition criteria 118 can be identified in various ways. For example, a retrieval process operating in accordance with the operational flow 100 may select the dividing criteria 118 itself based on predetermined criteria or based on information exchange with other retrieval processes, but is not limited to this example. Does not. Alternatively, partition criteria 118 can be obtained from another source, such as a control or coordination process or another search process that functions as a control or coordination process. Other options are available, and any method of identifying segmentation criteria 118 may be used.

Distance constraints provide some measure of the difference between two or more states. In one or more implementations, the distance constraint is related to one of the states and this one can be the reference state. In some implementations, the reference state can be the original state data provided to the search algorithm. In other implementations or in the same implementation at different time zones, the reference state can include some other state data. In some implementations of operation 160, using a distance constraint that is part of partition criterion 118, the reference state is, to a certain extent, similar to the identified new state or similar to existing state data 110. , Or similar to some other state data. In other implementations, distance constraints can again be used to assess whether the reference state differs from the identified new state to a certain degree.

In one exemplary implementation, distance constraints may be implemented using Hamming distance. The hamming distance is a measure of the number of changes that must be made to one state so that one state is equal to the other state. For example, assume that there are variables A, B, C, states a, b, c, and other states a, f, g. The hamming distance between these states is 2, because in order to make the first state equal to the first state, the value of B must be changed from b to f and the value of C must be changed from c to g. to be. In some cases, for example, when a hamming distance can be used with some local search algorithms, this hamming distance can represent the number of intermediate states between the start state and the end state.

In the same or other implementations, other operations or methods may be used as part of the implementation of the distance constraint. For example, in the case where state data is represented using a vector, the distance constraint may be, for example, projecting vectors onto another vector, projecting one vector or vectors onto a surface, and the like. It may be implemented using vector operations, but is not limited to this example.

Depending on the algorithm used for the distance constraint, the specific constraints of the distance constraint, and the choice of which states to use for the purpose of evaluating the distance constraint, various search investigation schemes can be obtained. For example, using operation flow 100, operation 160 may define a reference state as an initial state provided to the search algorithm, and identify only new states where the distance measurement from the reference state is less than the specified value. Distance constraints may be used, but are not limited to these examples. In another example, operation 160 may use other distance constraints to identify only new states where the distance measurement for the reference state is greater than some other value. In another example, operation 160 may use a distance constraint that specifies that only states having a distance measurement between two values or within a certain range are identified. Operation 160 may constrain or direct the search survey using distance measurements having many different states and values.

Selecting different values for the split reference parameters of the distance criterion and the reference state means that multiple instances of the search can be simultaneously (or sequentially) or time-shared for one computing resource using multiple individual computing resources. It may be useful if it occurs. For example, one instance of a search algorithm may be instructed to examine such states that are within a particular distance measure from a reference state. Another instance of the search algorithm may be instructed to examine such states that deviate from a particular distance measure from a reference state. One possible example system using one or more instances of a search algorithm is described below with reference to FIG. 2.

Referring to FIG. 2, an example and general system 200 is shown that examines a search space using one or more instances of search algorithms. One or more retrieval processes 210, 212, 214, and coordination or control processes 260 are included in system 200. In some implementations, one or more of the retrieval processes 210, 212, 214 can include coordination or control elements 270, 272, 274. In addition, one or more instances of split criteria 220, 222, 224, one or more instances of other data 230, 232, 234, one or more instances of output data 250, 252, 254, termination output data 280 Various data elements, including, are utilized by the system.

2 may be described with reference to other figures, it should be understood that the example system 200 is not intended to be limited to any particular figure or system of figures.

In one or more implementations, one or more of the search processes 210, 212, 214 retrieve a predetermined portion of the search space based on certain partitioning criteria 220, 222, 224 and other data 230, 232, 234. Implement an operational flow designed to For example, the retrieval process may execute the operation flow 100 described with reference to FIG. 1 or a predetermined operation flow in which the operation flow is appropriately modified. In other implementations, the one or more retrieval processes 210, 212, 214 can execute other operational flows. In some implementations, each search process can be executed on a particular individual computing resource, such as a computer processor with multiple cores, a particular computer processor, or a particular core in a particular computer in a distributed set of multiple computers. In other implementations, one or more search processes can share certain computing resources. Any number of search processes can be used. Different implementations may use different numbers of search processes depending on factors such as the complexity of the problem to be solved, the number of individually available computing resources, and the like.

The retrieval process that executes the operational flow may, in some implementations, utilize the given segmentation criteria 220, 222, 224 as well as other data 230, 232, 234.

As will be described in more detail below, the search topology for the entire system is determined in particular by the particular data being constrained or referenced by the partitioning criteria 220, 222, 224 and the number of search processes utilized. In some implementations, some search processes can have the same or similar partitioning criteria while other search processes can have different partitioning criteria. Similarly, various retrieval processes may use the same or different other data.

The segmentation criteria 220, 222, 224 may include both distance constraints and reference state information, as described above in detail. The segmentation criteria used by the search process may include a single set of distance constraints and reference states, or may include multiple sets of distance constraints and reference state information. In the latter case, which corresponds to multiple sets of distance constraint and reference state information, the search process may utilize some or all of the distance constraint and reference state information.

The other data 230, 232, 234 can include any additional data that is necessary or useful, in particular, for the search process to identify the output data 250, 252, 254. In some implementations, other data 230, 232, 234 can include abort criteria, but is not limited to this example. In some implementations, no other data 230, 232, 234 may be needed or required, and this data element may not be present or used by the system.

By using operational flows such as the provided segmentation criteria 220, 222, 224, other data 230, 232, 234, operational flow 100, the retrieval process 210, 212, 214 results in output data 250, 252. , 254). The output data 250, 252, 254 may include various information. This information may include any of the information generated as part of the operational flow executed as the operational flow 100, and / or other information. For example, output data 250, 252, 254 may include one or more states identified by a retrieval process, but is not limited to this example. In some cases, this one or more states may represent a suitable solution or solutions to the problem to be solved, at least, such as a solution found by a particular search process that uses the provided partitioning criteria and may also use other data. In the same or other implementations, the output data may include multiple states, suitability or adaptability of the states returned or examined, and / or other data.

The output data 250, 252, 254 may then be used by the adjustment or control process 260, which may identify the predetermined end output data 280 or may use a new set of partitioning criteria 220, 222, 224 and other data 230, 232, 234 can also be identified, which in turn are used to perform further investigation of the search space by using search processes 210, 212, 214.

The termination output data 280 may include various information. In some example implementations, the termination output data 280 is identified by the retrieval process 210, 212, 214 and is a user of the system 200 by the retrieval processes and / or the coordination or control process 260. It may include one or more states determined to be related to. For example, the termination output data 280 may include one most suitable output state found by any of the retrieval processes across all iterations of the operational flows used by the system 200. However, it is not limited to this example. In other example implementations, the termination output data can include multiple output states. In addition, the termination output data may include other information useful to the user of the system, such as detailed result data 112 generated by the operational flow 100, but is not limited to this example. In addition, the termination output data may include, but is not limited to, for example, information generated by an adjustment or control process, including an aggregate summary of a search survey process, or other data.

The adjustment or control process 260 can use various methods to stop execution and determine whether to return the predetermined exit output data 280. In some implementations, the adjustment or control process can return termination output data if the output data 250, 252, 254 falls within a predetermined threshold provided by the user of the system 200. For example, the process may return the end output data only when the end output data exceeds a predetermined value or only when the result data 112 of the end output data falls within a specific range. In the same or other implementations, the coordination or control process 260 may return end output data when executing the operational flows used by the retrieval processes 210, 212, 214 a predetermined number of times. For example, the process may execute operational flows for a fixed number of iterations or only for a predetermined amount of time, and then return some exit output data 280.

If the adjustment or control process 260 does not return the exit output data 280, this process will also identify a new set of partition criteria 220, 222, 224 and other data 230, 232, 234 instead. These can then be used to perform further investigation of the search space by reusing search processes 210, 212, 214. The manner in which the coordination or control process 260 identifies the nature of this data and / or the data itself can vary widely.

The new segmentation criteria 220, 222, 224 and the new other data 230, 232, 234 may be the same for some search processes or may be different for some or all search processes. For example, in one implementation, the adjustment or control process 260 may use the same reference state data and other data and change only the distance constraint data. In other implementations, the adjustment or control process 260 can change both the reference state and distance constraints for some or all of the search processes. Some examples of search schemes resulting from different partitioning criteria are described below with reference to FIGS. 4 through 9.

The adjustment or control process 260 can use any method to identify new segmentation criteria 220, 222, 224 and new other data 230, 232, 234. In one or more example implementations, the coordination or control process 260 can identify only one reference state for all retrieval processes and from previous iterations as provided in the output data 250, 252, 254. You can identify the state by identifying the most appropriate state. In other implementations, this process can identify the reference state using some other means, or can identify different reference states for different search processes. Similarly, the coordination or control process 260 may use different distance constraints for each search process, for example, dividing the search space into different partitions and directing the survey in different ways. can do.

In some implementations, the identification of the output data, the segmentation criteria, and / or the new states, for investigation, is performed at one or more adjustment or control elements 270, 272, 274 instead of one adjustment or control process 260. Can be performed. In some implementations, these coordination or control elements can relate to one or more search processes 210, 212, 214. One possible implementation of the operational flow of identifying exit output data and new states for investigation using elements such as one or more adjustment or control elements is described in more detail below with reference to FIG. 3.

In some of these implementations, only one coordination or control element 270 may be required to direct and coordinate or control the entire search investigation process of the plurality of search processes 210, 212, 214, but is limited to this example. It doesn't work. For example, the first search process to start may detect that it is the first search process and may control any search processes that are subsequently started. In other implementations, each retrieval process 210, 212, 214 can have its own coordination or control process 270, 272, 274, but is not limited to this example. In such a system, the coordination or control processes may communicate with each other to determine when to stop the full search investigation and identify new states to investigate, new segmentation criteria to be used, and the like. In still other implementations, some but not all of the retrieval processes can include coordination or control elements.

Referring now to FIG. 3, an exemplary general operational flow 300 is shown, including various operations that may be performed using search algorithms to examine a search space, without a centralized partitioning agent or logic. Exemplary operational flow 300 directs its search by examining the search space and, in some implementations, possibly communicating with other systems executing an operational flow, such as operational flow 300.

3 may be described with reference to other figures, it should be understood that the illustrative operational flow 300 is not intended to be limited to anything related to any particular figure or system of figures. Also, while the example operational flow 300 indicates a particular order of execution of the operations, it is to be understood that in one or more other implementations, the operations may have a different order. In addition, some of the data and steps illustrated in the example operational flow 300 may not be required and may be omitted in some implementations. Finally, while this flow of operations involves a number of individual steps, it should be appreciated that in some circumstances some of these operations may be combined and executed simultaneously.

The system 200 of FIG. 2 illustrates one use for the example operational flow 100, examining the search space for possible solutions or solutions to problems. One implementation of the system 200 uses the coordination or control process 260 to direct the overall operation of the retrieval processes 210, 212, 214 executing the action flow 100, but the coordination or control process Using a single splitting agent, such as 260, is not necessary in all implementations.

In other implementations, some other method may be used to perform derivation of search processes 210, 212, 214. For example, the operational flows executed by the retrieval processes may self identify the exit output data and new states to examine and may not require any centralized partitioning logic. For example, each retrieval process 210, 212, 214 can execute an operational flow like the retrieval process of the operational flow 300, but is not limited to this example. This may occur in particular by using one or more adjustment or control elements 270, 272, 274, or may occur through some other element, or may occur within the search process itself. In such or other implementations, the exchange / split operation 330 can achieve some or all of the same functionality provided by the coordination or control process 260. In one or more implementations, each of the plurality of retrieval processes may execute operation flow 300 concurrently, exchange data or communicate to further direct the survey.

As shown, one implementation of operation 310 examines the search space. The investigation may be accomplished using an operational flow such as that of the operational flow 100 of FIG. 1 described above, or using some other operational flow. In the example using the operational flow 100, after the operational flow 300 proceeds from the operation 310, any of the data used or generated by the operational flow 100 is a subsequent operation in the operational flow 300. Are available for these. This data includes, but is not limited to, state data 110, and result data 112, including states and initial states that were examined during a search.

In one implementation of operation 320, the operation flow 300 determines whether this instance of itself should stop execution. If it is determined in operation 320 that this instance of the operation flow 300 should be stopped, the operation flow 300 ends (“Yes” branch, operation 320). If it is determined that operation flow 300 should not be interrupted, operation flow proceeds to operation 330 ("No" branch, operation 320), as described below.

The criteria used by operation 320 to determine whether the operational flow should be interrupted can vary widely. For example, in one implementation, operation 320 may determine execution halt if one or more of the results returned by execution of operation 310 fall within an acceptable range, but is not limited to this example. In the same or other implementations, operation 320 may cease execution of the operation flow when operation flow 300 has been executed for a predetermined number of set iterations or for a predetermined amount of time, but likewise is not limited to this example. Do not. In another implementation or implementations, operation 320 may stop execution of operation flow 300 if any other operation flow finds a suitable solution.

In some implementations, operation 320 can stop execution of only a particular instance of operation flow 300. In the same or other implementations, operation 320 may directly or indirectly suspend execution of one or more other instances of concurrently executing operation flow 300. For example, in some implementations, if this particular instance of operation flow 300 identifies an acceptable solution to the problem, operation 320 may cease execution and all other operations of operation flow 300 The instance may also communicate that it should stop running.

If operation flow 300 is not interrupted by operation 320, operation flow proceeds to operation 330. In one or more implementations of operation 330, data may be exchanged with other concurrently running instances of operation flow 300, and a manner of dividing or further examining the search space may be determined such that It may be communicated between one or more of the instances.

The manner in which the data exchanged and the search space are partitioned can vary widely. In one or more implementations, for example, a particular instance of the operational flow 300 may identify a particular partitioning criterion for the next iteration, but is not limited to this example. This instance of the operational flow 300 may then communicate, to other instances of the operational flow 300, all or part of the selected partitioning criteria and also communicate instructions for other partitions to retrieve. have. Thereafter, other instances of the operational flow 300 may perform a search according to the instructions provided by the first instance of the operational flow.

In one or more implementations of operation 300, each instance of operation flow 300 may identify desirable values for the segmentation criteria, and then communicate these values to other instances executing operation flow 300. do. Thereafter, some logic in each instance of operation 330 may evaluate how to compare the initially selected particular segmentation criteria to the particular criteria selected by other instances of the operational flow. In some implementations, operation 330 can identify a new partition criterion and then communicate this new partition criterion to other instances executing operation 330. In some implementations, this process of selecting candidate splitting criteria, communicating with other instances of operation flow 300, and updating the selected splitting criteria can be, for example, until the selected splitting criteria are the same or in a different perspective. Can be run any number of times until it is appropriate.

Similar to the splitting criteria described with reference to FIGS. 1 and 2, in some implementations, the splitting criteria used by one instance of the operational flow 300 is the splitting criteria used by other instances of the operational flow 300. It may be the same as the standard. In the same or other implementations, the segmentation criteria used by one instance of the operational flow 300 may be different from the segmentation criteria used by another instance of the operational flow 300. In addition, the splitting criteria may be different due to different reference conditions, different distance constraints, or both. Likewise, any other data used by one instance of operation flow 300 may be the same or different from other data used by other instances of operation flow 300.

It is to be understood that the manner in which operation 300 identifies the segmentation criteria to be used for other iterations of the operational flow can vary widely and is not limited in any way by the examples described above.

4 to 9, the search topology obtained by using various partitioning criteria having a plurality of search processes is examined. As discussed above, the segmentation criteria may include reference conditions and distance constraints. The search topology may be generated by identifying the same or different values for reference conditions and distance constraints for multiple search processes. In some cases, the topologies illustrated with respect to FIGS. 4-9 may be considered relatively static, but they may be combined such that the search topology may evolve over time or change throughout the entire search.

In each of FIGS. 4-9, a suitable implementation of the manner in which the illustrated topology can be implemented is the system and operational flow described above with respect to FIGS. For example, a topology illustrating two search spaces may utilize the search process 210 and the search process 212 described above with reference to FIG. 2. Any other embodiment or variations thereof is also available. In one implementation of the operational flow 100 described with reference to FIG. 1, operation 160 may constrain the new states that it identifies using the segmentation criteria described below.

Although FIGS. 4-9 may be described with reference to other figures, it is to be understood that the illustrations of FIGS. 4-9 are not intended to be limited to anything related to the system or other content of any particular figure or figures. In addition, since FIGS. 4-9 illustrate a general representation of an exemplary data set for convenience of description, there is no further inference from the nature or configuration of the illustrated shape or from the identification of data values. The example described herein may be represented using two-dimensional space, which may be a suitable representation of a problem space modeled using two variables, but the problem space and search investigation methods may be any number of variables or any Please understand that it can be expressed using other expressions. For example, the methods of problem space and search investigation can be expressed in part or in whole using a hyperplan generated by a Cartesian product of variables used to represent states.

Referring now to FIG. 4, an illustrative and general representation of a system 400 illustrating one of many possible search topologies by dividing a search space into two search spaces is shown. In system 400, an entire search space 410, a first search space 420, a second search space 430, a reference state 424, and a distance constraint 426 are shown. Note that each of the subsequent search topologies described below with reference to FIGS. 5-9 may include some similar elements and that the description of these similar elements may not be repeated for each example.

The search topology exemplified with reference to FIG. 4 may include the entire search space 410 representing the entire problem area that can be searched for a solution, the search space to be processed by the first search process and the second search process. This can occur using different partitioning criteria divided into search spaces. In the example system 400, the reference state for both the first search space and the second search space may be the same and may be represented by the reference state 424. The distance constraint for the first search process may be "d <D" and may be represented by distance constraint 426. In this context, “d <D” may indicate that the distance measurement for the states retrieved by the first search process should be less than the predetermined value D from 424 from the reference state. On the other hand, the second search process may have a distance constraint of "d> = D", which, in this example, determines the states considered by the second search process against the same reference state 424 again. Can be constrained to states having a distance measurement equal to or greater than the same predetermined value D.

According to these selections of partitioning criteria, the first search space 420 can be searched by the first search process and the second search space 430 can be searched by the second search process.

There are many possible uses of the search topology as shown with respect to FIG. 4, but one possible use is the second while using the first search space 420 to search for a specific area of the entire search space 410 in detail. The search space 430 may be used to search the rest of the entire search space 410 in less detail.

Referring now to FIG. 5, an exemplary representation of a system 500 is illustrated that illustrates how a search can be changed over time by using new segmentation criteria. This particular example illustrates one possible modification of the search topology from the search topology described above with respect to FIG. 4. The system 500 includes an entire search space 410, a first search space 520, a second search space 530, an old first search space 420 (same as the first search space 420), and Reference state 524 is shown.

The search topology illustrated with respect to FIG. 5 may in this case illustrate one possible way of instructing and changing the search using two search processes over time from the search topology illustrated with respect to FIG. 4. Similar to the example system 400, in the example system 500, the reference state for both the first search process and the second search process is the same, that is, the reference state 524, while the distance constraints are Are different. That is, "d <D" and "d> = D". However, because the reference state 524 has changed from the reference state 424 in the system 400, the first and second search processes now search for different portions of the overall search space 410.

Perhaps by periodically changing the partitioning criteria in a manner similar to that illustrated for FIG. 5 or in some other manner, the entire search may proceed to find more and more suitable solutions. For example, in one implementation illustrated with respect to FIGS. 4 and 5 where one search process performs a refinement while the other performs a more general search, the search processes sometimes provide their most suitable solutions. Exchange and comparison is possible, but is not limited to these examples. In one implementation, this may perform a more detailed search for possible more suitable search spaces whose location is known by a general search.

In one implementation, the change in partitioning criteria may be performed by the adjustment or control process 260 or adjustment or control elements 270, 272, 274 described above with respect to FIG. "May be related to the execution result of operation 330. In other implementations, the change in partitioning criteria can be related to any other system element or operation.

Referring now to FIG. 6, an illustrative, general representation of a system 600 is illustrated that illustrates one of the many possible separations that divide a search space into three search spaces. In system 600, an entire search space 610, a first search space 620, a second search space 630, a third search space 640, and a reference state 624 are shown.

The search topology illustrated with respect to FIG. 6 may occur by dividing the entire search space 610 into three search spaces, each processed by one search process, using different partitioning criteria. The first search process may search the first search space 620, the second search process may search the second search space 630, and the third search process may search the third search space 640. can do.

In the example system 600, the reference state for each search process may be the same, that is, the reference state 624. The distance constraint for the first search process may be "d <D1", the distance constraint for the second search process may be "D2> d> = D1", and the distance constraint for the third search process May be "d> = D3". The distance constraints (distance constraint 642, distance constraint 646, respectively) for the first search space 620 and the third search space 640 are distance measures for the states they searched for. 4 and 5 in that it is possible to define a search space that is smaller than a predetermined value, such as D1 (for the first search space 620) or more than a predetermined value, such as D3 (for the second search space 640). Similar to the distance constraints described above for. The dividing criterion for the second search space 630 uses " D2 > d > = D1 ", which is a distance constraint 644 that includes two specific values, thus in some respects the first search space ( Search for states that are between the distance constraints of 620 and third search space 640.

Depending on how D2 and D3 are selected, it is possible to carefully create overlapping regions between search spaces. For example, in some cases if D2> D3, an overlap region illustrated at 632 may be generated. On the other hand, if D2 = D3, no overlapping region is generated. Illustrate overlap 632 that may occur to illustrate the flexibility available by the segmentation mechanism described herein.

Referring now to FIG. 7, an exemplary and general representation of a system 700 is illustrated that illustrates that reference states are not equal and search spaces are divided into two search spaces that do not include all of the search spaces. In system 700, an entire search space 710, a first search space 720, a second search space 730, a first spatial reference state 724, and a second spatial reference state 734 are shown. .

The search topology illustrated with respect to FIG. 7 includes a first search space 720 and a second search space 730 where the entire search space 710 is processed by one search process for each search space using different partitioning criteria. By dividing by). Unlike some of the examples described above, system 700 illustrates a search topology in which the reference states that are part of the partitioning criteria are not identical. In this example, the reference state for the first search space is illustrated by the first spatial reference state 724 while the reference state for the second search space is illustrated by the second spatial reference state 734. In addition, the distance constraints that are part of each division criterion are also different in both the reference state for them and the distance constraints themselves. The distance constraint 740 for the first search space 720 is "d <D1" while the distance constraint 742 for the second search space 730 is "d <D2". As a result, the size illustrating the second search space 730 is wider than the size of the first search space 720, which may mean that in some examples, D2 is greater than D1, although not required.

Finally, the system 700 also illustrates that not all of the entire search space 710 is searched. The example system 700 has only two search processes, each of which searches for a detail area around its own reference state. Some of the entire search space 710, which falls outside of these detailed areas, is not searched. While useful in some cases, covering the entire search space by at least one search space is not a requirement. Further, in some implementations, additional iterations, in which one or more of the iteration criteria can be changed, can include some or all of the search space that is not currently being searched.

Referring now to FIG. 8, an illustrative and general representation of a system 800 is illustrated illustrating separation into two search spaces where the reference states are not the same and the final search spaces overlap. System 800 includes an entire search space 810, a first search space 820, a second search space 830, a first spatial reference state 824, a second spatial reference state 834, and a shared Search space 850 is shown.

As with some of the examples described above, the search topology illustrated with respect to FIG. 6 includes a first search space 820 and a second search space 830 where the entire search space 810 is processed by one search process for each search space. Can be derived from different partitioning criteria. As shown, the reference states for the two search spaces are different, that is, the first search space uses the first spatial search state 824 while the second search space uses the second spatial search state 834. .

In accordance with certain choices for partitioning criteria in this example system, a portion of the entire search space 810 is searched by all search processes. Some of these overlaps are illustrated by shared search space 850. As can be seen in this example, the search processes do not always need to look into unequal search spaces.

Referring now to FIG. 9, an exemplary and general representation of a system 900 that illustrates the separation into three search spaces in which a third search space is defined using partitioning criteria including reference states and distance constraints. Is shown. The system 900 includes an overall search space 910, a second search space 920, a second search space 930, a third search space 940, a first space reference state 924, and a second space reference. State 934 is shown.

The first search space 920 and the second search space 930 are defined using a split criterion similar to the split criterion described above with reference to, for example, FIG. 7. These spaces, in this example, are different reference states, the first spatial reference state, as well as distance constraints that define a search space that includes states in the distance measurement specified by the set value, respectively D1 and D2. 924 and the second spatial reference state 934 are used, respectively.

On the other hand, the third search space 940 is defined in a different way from some other search spaces defined. The third search space 940 is defined using a split criterion that includes two reference states and two distance constraints. In this example, the third search space 940 is determined using the distance constraint "d1> = D1" determined from the first space reference state 924 and from the second space reference state 934. Defined using a distance constraint of "d2> = D2". In this example, this partitioning criterion constrains the states considered by the third search process to only those states that are not part of the first search space 920 or parts of the second search space 930.

Referring now to FIG. 10, one possible basic implementation of a computing device is shown. 10 and related descriptions are intended to provide a brief and general description of an example computing environment that may implement the various techniques described herein. Although not required, the techniques herein describe computer executable instructions, such as program modules, executed by a controller, processor, personal computer, or other computing device, such as computing device 1000 illustrated in FIG. 10. It is described at least in part in the general context.

Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. Tasks performed by the program modules are described above with reference to block diagrams and operational flow diagrams.

Skilled artisans may implement the description, block diagrams, flow diagrams in the form of computer executable instructions that may be embodied in one or more forms of computer readable media. As used herein, a computer readable medium can be any medium that can store or implement information encoded in a form that can be accessed and understood by a computer. Typical types of computer readable media include, but are not limited to, volatile and nonvolatile memory, data storage devices, including removable and / or non-removable media, and communication media.

Communication media embody computer readable information on modulated data signals, such as carrier waves or other transport mechanisms, and include all information delivery media. The term " modulated data signal " means a signal that has one or more of its characteristics set or changed to encode information in the signal. By way of example, communication media includes, but is not limited to, wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, or other wireless media.

The computing device 1000 of FIG. 10, in its most basic configuration, includes at least one processing device 1002 and a memory 1004. Depending on the exact configuration and type of computing device, memory 1004 may be volatile (such as RAM), nonvolatile (such as ROM, flash memory, etc.), or some combination of both. This most basic configuration is illustrated by dashed line 1006 in FIG. In addition, computing device 1000 may have additional features / functionality. For example, computing device 1000 may include, but is not limited to, additional storage (removable and / or non-removable), including magnetic or optical disks or tapes. This additional reservoir is illustrated by the removable reservoir 1008 and the non-removable reservoir 1010 in FIG. 10.

Computing device 1000 may include one or more communication connections 1012 that enable computing device 1000 to communicate with other devices. The computing device 1000 may have one or more input devices 1014, such as a keyboard, mouse, pen, voice input device, touch input device, or the like. One or more output devices 1016, such as a display, speaker, printer, or the like, may also be included in the computing device 1000.

Those skilled in the art will appreciate that the techniques described herein may be implemented with computing devices other than the computing device 1000 illustrated in FIG. 10. For example, the techniques described herein are hand-held devices, multiprocessor systems, microprocessor-based or programmable consumer electronics. The present invention may be implemented in a network PC, a minicomputer, a mainframe computer, or the like, but is not limited thereto.

The techniques described herein may be implemented in a distributed computing environment where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media.

Although the foregoing techniques have been described as being implemented in software, alternatively, the techniques described above may be implemented in part or in whole as hardware, firmware, or various combinations of software, hardware, and / or firmware. I understand that.

Claims (20)

a) for each process of performing a search in a search space, identifying at least one reference search state in the search space, b) for each process of performing a search in the search space, identifying at least one distance constraint to be used with the at least one reference search state to select current search states in the search space. How to include. The method of claim 1, Each process of searching in the search space, c) identifying a current search state based on the at least one distance constraint and the at least one reference state; d) evaluating a model using the current search state; e) selecting a new state based on the at least one distance constraint and the at least one reference state, and using the new state as the current search state; f) repeating steps d) to e) until the stopping criteria are met How to perform a method that includes. The method of claim 2, At least two processes searching the search space. The method of claim 3, a) the at least one reference search state in the search space is the same for each process performing a search in the search space, b) the at least one distance constraint for each process of searching in the search space is selected to divide the search space into regions. The method of claim 4, wherein At least two regions at least partially overlap. The method of claim 4, wherein At least two regions are non-overlapping. The method of claim 3, a) at least one distance constraint is different for at least one process of searching in the search space and other processes of searching in the search space. The method of claim 3, a) at least one reference search state in the search space is different for at least one process for searching in the search space and other processes for searching in the search space. The method of claim 8, And said at least one distance constraint for each process of performing a search in said search space is selected to partition said search space into regions. The method of claim 9, At least two regions at least partially overlap. The method of claim 9, At least two regions are non-overlapping. The method of claim 2, Each of the at least two processes execute the method of claim 2, At least one of the at least two processes further exchanges its at least one reference search state with at least another processor of the at least two processes. The method of claim 2, Each process of searching in the search space, a) identify its own at least one criteria search state, b) identify their own at least one distance constraint, c) each process performing a search in the search space further executes a method, The method is i) exchanging information containing at least one reference state of its own with at least one of the at least two processes; ii) performing steps c) to f) of claim 2, If the abort criteria of step f) are met, iii) identifying a new at least one reference state and using the new at least one reference state as the at least one reference state; iv) identify a new at least one distance constraint to be used with the at least one reference search state to select current search states in the search space, and apply the new at least one distance constraint to the at least one Using as a distance constraint, v) repeating steps i) to (iv) until a global stop criterion is reached; How to include. The method of claim 2, g) identifying at least one reference search state within the search space by an adjustment process for each process performing a search in the search space; h) identifying, by the adjustment process for each process performing a search in the search space, at least one distance constraint to be used with the at least one reference search state to select current search states in the search space. Steps, i) each process of searching in the search space performing steps c) to f) of claim 2, If the abort criteria of step f) are met, j) identifying a new at least one reference state by the reconciliation process and using the new at least one reference state as the at least one reference state by processes for searching the search space; k) identifying, by the adjustment process, a new at least one distance constraint to be used with the at least one reference search state to select current search states in the search space, and wherein the new at least one distance constraint Using a condition as said at least one distance constraint; l) repeating steps i) to k) until a global abort criterion is reached; How to do more. A computer readable medium comprising executable instructions for performing the method of claim 1. A computer readable medium having encoded executable instructions, the computer readable medium comprising: a) means for identifying at least one segmentation criterion for partitioning the search space into a plurality of regions, Each of the plurality of areas represents a portion of the search space to be searched by a search process, Wherein each of the at least one partitioning criterion comprises at least one reference search state and at least one distance constraint. The method of claim 16, Means for identifying a current search state based on the at least one partitioning criterion. As a system for searching a search space, a) a plurality of computing resources, b) a plurality of retrieval processes running on at least one of the plurality of computing resources; c) at least one control process identifying at least one segmentation criterion, wherein the segmentation criterion comprises at least one distance measure for a reference condition; Including, d) each of the plurality of search processes is configured to select a current search state based on the at least one segmentation criterion. The method of claim 18, At least one search process is the control process, The control process identifies at least one segmentation criterion for each of the plurality of search processes, And the control process is executed on at least one computing resource of the plurality of computing resources. The method of claim 18, And the distance measure is a Hamming distance.

Images